Digital Watermarking
Multimedia Security
What is the Watermark ?
• Paper Watermark
– the technique of impressing into
the paper a form, image, or text
• to make forgery more difficult
• to record the manufacturer’s
trademark
2
Digital Watermark
• A digital watermark
– a digital signal or pattern imposed on a digital document ( text,
graphics, multimedia presentations )
• visible watermark
– the more obvious means of discouraging unauthorized use by
reducing the commercial value of a document
• invisible watermark
– the watermark is imperceptible to the human eye
– when the ownership of data is in question, the watermark will then
be extracted to characterize the ownership
3
Visible Watermarking
• G. W. Braudaway, K. A. Magerlein and F. Mintzer, “Protecting publiclyavailable images with a visible watermark,” SPIE, vol. 2659, 1996, pp.
126-133.
4
Invisible Watermark
• Motivation
– The distribution of digital media is becoming faster, easier and
requiring less effort to make exact copies
• How to protect the intellectual property?
• Conventional approaches
– In analog world
• signature, steel seal, embossed portrait, copyright label...
– In digital world: cryptology
M
E
encryption key
D
M
decryption key
5
Cryptology vs. Watermarking
• Cryptology
Once the data is decrypted, subsequent retransmission
or dissemination is not encrypted
• Watermarking
– Copyright information is hidden into digital data itself
– Not restrict to access the data
– Its objective is to permanently and unalterably reside in
the data
6
Watermarking Requirements
• Imperceptible
• Undeletable
• Statistically undetectable
• Robustness
– resistant to lossy data compression
– resistant to signal manipulation and
processing operation
• Unambiguous
7
Watermarked Image
Transmission
Lossy
Compression
Geometric
Distortions
Signal
Processing
D/A - A/D
Conversion
Typical Distortions or Intentional Tampering
Transmission
Corrupted Watermarked Image
8
Watermark Embedding
• Making the watermark robustness is not trivial
– with complete knowledge
• any watermark can theoretically be removed
– with partial knowledge
• the removal may interfere with the viewing of the data
• the effort of removal is greater than the value of the data
• Challenges from data compression
– Whatever hole one may find to fill with watermark is likely to be
eliminated by data compression
9
Watermarking for Text
J.T. Brassil,”Electronic Marking and Identification Techniques
Discourage Document Copying,” IEEE JSAC, vol 13, no.8, Oct, 1995.
to
• Line-Shift Coding
• Word-Shift Coding
•
Feature Coding
10
Watermarking for Images & Videos
• Watermarking in
spatial domain
transform domain
• Watermarking in
raw data
compressed data
• Watermarking with
random number
visually recognizable pattern
• Detection/extraction
with
without
the original data
11
LSB Flipping Method
S. Walton, “Image Authentication for a Slippery New Age,” Dr. Dobb’s
Journal, pp. 18-26, April 1995.
0
1
2
3
4
Bit 1
0
Bit 0
Watermark 1
1
Bit 1
Bit 2
Watermark 2
2
Bit 3
Bit 0
5
Bit 2
3
Bit 3 4
5
0
1
2
3
4
5
55
73
123
123
205
120
121
130
123
71
71
70
72
147
199
67
68
68
72
73
123
117
123
70
70
69
112
67
140
158
159
195
123
123
133
142
178
120
123
122
150
71
70
70
0
1
2
3
4
5
• Generate the random walk sequence for each watermark (e.g.. 00112)
• Force the LSB to match the watermark bit

This works will not survive any modification
12
Spread Spectrum Method
I. J. Cox,”Secure Spread Spectrum Watermarking for Multimedia,” IEEE
Trans. IP, vol 6, no.12, pp. 1673-1687, Dec. 1997.
• Spread spectrum coding of a watermark
– frequency domain of the image
– watermark
communication channel
the signal transmitted through the channel
the watermark W = w1,...,wn
each wi is chosen according to zero-mean Gaussian Distribution
the image X is transform by full-frame DCT
n highest magnitude coefficients (except DC) are chosen: y1,...,yn
Embedding: y’i = yi + a wi
Extracting:
wi = (yi* - yi) / a
similarity = correlation (W,W *)
13
Spread Spectrum Method (cont.)
Original
Image
FFT/DCT
FFT/DCT
Watermark
Received
Image
Original
Image
Determine Perceptually
Significant Regions
－
Original
Watermark
Extracted
Watermark
Inverse FFT/DCT
Watermarked
Image
Similar
14
Spread Spectrum Method (cont.)
• Watermark detector
Watermark detector response to 1000 randomly generated watermarks
15
Perceptually Masking Method
M. D. Swanson,”Transparent Robust Image Watermarking,” Proc. ICIP, pp.
211-214, Sep. 1996.
Original
Image
F
DCT
+
F’
W
Perceptual
Analysis
m-sequence
x
IDCT
Watermarked
Image
Spatial
Masking
DCT
 Detection
H 0 : X = F* - F = N
H 1 : X = F* - F = W * + N
the hypothesis decision is obtained by
similarity = correlation (X,W)
16
Perceptually Masking Method (cont.)
• Frame from “pingpong”
Similarity value
• The watermark
17
Digimarc Watermarking
• A commercial watermarking software
– http://www.digimarc.com
Creator ID
100024
18
Watermarking for Audio
• Phase Coding
– Inserting the watermark by modifying the phase of each
frequency component
• Spread Spectrum Method
– The watermark code is spread over the available frequency band,
and then attenuated and added as additive random noise
• Perceptual Method
– The watermark is generated by filtering a PN-sequence with a
filter that approximates the frequency masking characteristics of
HAS
– Weighting the watermark in the time domain to account for
temporal masking
19
Watermarking for Audio (cont.)
L. Boney, “Digital Watermarks for Audio Signals, “ IEEE Proc.
Multimedia’96, pp. 473-480, 1996.
• Watermark generator
Audio
Signal
Hanning
Window
FFT
Frequency
Masking
Information
Quantization
Scale
Factor
PN
sequence
w+
original
signal
w
M(w)
×
×
+
Extract
Envelop
Filtered PN-sequence
20
Watermarking for Audio (cont.)
• Watermark detection
21
Watermarking for Polygonal Models
R. Ohbuchi, “Watermarking 3-D Polygonal Models,” ACM Multimedia 97,
pp. 261-272, 1997.
• 3D models watermarking
– vertex coordinates
– vertex topology (connectivity)
Embedded pattern
Simplified polygonal
22
Limitations of Watermarking
S. Craver, “Can Invisible Watermarks Resolve Rightful Ownerships?,” IBM
Research Report, July, 1996.
• Basic watermarking steps
Image I
I’
E
Watermark W
Test image J
Image I
D
W’
C
y/n ?
Watermark W
23
Limitations of Watermarking (cont.)
• Counterfeit
Watermarked image I’
Counterfeit image I*
Dinv
W*
I’
D
I
I’
C
y/n ?
C
y/n ?
Watermark W
D
I*
W
W*
Watermark W*
24
Digital watermark
Visible watermark
Invisible watermark
Random sequence
watermark
Visually recognizable
watermark
Watermark :
ID number (random number)
Visually recognizable pattern
Verification :
Quantitative measurement of
the detection
Extracted pattern &
Quantitative measurement
25